Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method for estimating body orientation, the method comprising: executing a region extraction processing that includes extracting a person region from a plurality of images, the person region corresponding to a region of a person, the plurality of images including an image taken at a first time point and an image taken at a second time point after the first time point; executing an identification processing that includes identifying a first body part and a second body part in the person region from each image at the first time point and the second time point, the first body part being a left shoulder of the person in the person region, the second body part being a right shoulder of the person in the person region; executing a calculation processing that includes calculating a first moving amount and a second moving amount, the first moving amount indicating a moving amount between a position of the first body part at the first time point and a position of the first body part at the second time point, the second moving amount indicating a moving amount between a position of the second body part at the first time point and a position of the second body part at the second time point; and executing an estimation processing that includes: in response to detecting that the first moving amount is larger than the second moving amount, estimating body orientation of the person in the person region at the second time point by using first relation information and a difference between the first moving amount and the second moving amount, the first relation information indicating a relationship between the difference and a first change amount of the body orientation when the person rotates clockwise; and in response to detecting that the second moving amount is larger than the first moving amount, estimating the body orientation of the person in the person region at the second time by using second relation information and the difference, the second relation information indicating a relationship between the difference and a second change amount of the body orientation when the person rotates counter-clockwise.
The invention relates to a method for estimating body orientation from a sequence of images. The method addresses the challenge of accurately determining a person's body orientation over time using visual data. The process begins by extracting a person's region from multiple images captured at different time points. Within this region, the system identifies key body parts, specifically the left and right shoulders. The method then calculates the movement of these shoulders between the first and second time points, determining the displacement of each shoulder. Based on the relative movement of the shoulders, the system estimates the body orientation. If the left shoulder moves more than the right, the method uses predefined relation information to estimate the orientation change, assuming a clockwise rotation. Conversely, if the right shoulder moves more, it applies different relation information to estimate a counter-clockwise rotation. The method leverages the difference in shoulder movement to infer rotational changes in body orientation, providing a dynamic estimation of posture over time. This approach is useful in applications requiring real-time body tracking, such as motion analysis, surveillance, or human-computer interaction.
2. The method according to claim 1 , wherein the first relation information and the second relation information are set for each body orientation at the first time point, the estimation processing is configured to: in response to detecting that the first moving amount is larger than the second moving amount, identify the first relation information corresponding to the estimated body orientation at the first time point; and in response to detecting that the second moving amount is larger than the first moving amount, identify the second relation information corresponding to the estimated body orientation at the first time point.
This invention relates to a method for estimating body orientation and movement using relation information. The method addresses the challenge of accurately determining body orientation and movement in scenarios where different body parts or sensors may exhibit varying levels of movement. The method involves tracking the movement of at least two body parts or sensors over time, comparing their movement amounts, and selecting relation information based on which body part or sensor exhibits greater movement. The relation information is predefined for different body orientations and is used to refine the estimation of body orientation. When the movement of a first body part or sensor is greater than that of a second, the method selects the first relation information corresponding to the estimated body orientation. Conversely, if the second body part or sensor moves more, the second relation information is selected. This approach ensures that the most reliable movement data is used to improve the accuracy of body orientation estimation, particularly in dynamic or ambiguous scenarios. The method is applicable in fields such as motion tracking, wearable technology, and human-computer interaction, where precise body orientation estimation is critical.
3. The method according to claim 1 , wherein the identification processing is configured to use feature amounts at a plurality of feature points extracted from the person region in order to identify the first body part and the second body part.
The invention relates to a method for identifying body parts of a person in an image or video. The method addresses the challenge of accurately distinguishing between different body parts, such as arms, legs, or torso, in images where the person may be in various poses or occluded. The method involves extracting feature points from a detected person region in the image, where these feature points represent key locations or characteristics of the body. Feature amounts, which are quantitative measurements derived from these feature points, are then used to identify and classify specific body parts. For example, the feature amounts may include spatial coordinates, angles, or other geometric properties of the feature points. By analyzing these feature amounts, the method can distinguish between the first body part (e.g., an arm) and the second body part (e.g., a leg). The method may also involve preprocessing steps to enhance the accuracy of feature extraction, such as noise reduction or contrast adjustment. The overall goal is to provide a robust and automated way to identify body parts for applications like motion analysis, gesture recognition, or human-computer interaction.
4. The method according to claim 3 , further comprising: storing, as first feature points associated with the first time point, second feature points identified from the person region at the second time point; and storing the estimated body orientation at the second time as the body orientation at the first time.
This invention relates to a method for tracking human body orientation over time using feature points extracted from video frames. The method addresses the challenge of accurately estimating and maintaining consistent body orientation data when tracking a person across multiple frames, particularly in scenarios where the person may move or change position. The method involves capturing video frames of a person at different time points, including at least a first time point and a second time point. A person region is identified in each frame, and feature points are extracted from these regions. At the first time point, feature points are identified and stored as first feature points. At the second time point, second feature points are identified from the person region and stored as first feature points associated with the first time point. Additionally, the estimated body orientation at the second time is stored as the body orientation at the first time. This approach ensures that feature points from subsequent frames are linked back to the initial time point, allowing for consistent tracking of body orientation over time. The method helps maintain accuracy in orientation estimation even when the person's position or appearance changes between frames. The stored feature points and orientation data can be used for applications such as motion analysis, activity recognition, or human-computer interaction.
5. The method according to claim 1 , wherein the region extraction processing is configured to extract a plurality of person regions in an image, the identification processing is configured to identify, for each of the plurality of person regions, the first body part in the person region and the second body part in the person region, the calculation processing is configured to calculate, for each of the plurality of person regions, the first moving amount and the second moving amount, and the estimation processing is configured to estimate, for each of the plurality of person regions, the body orientation of the person in the person region at the second time.
6. An apparatus for estimating body orientation, the apparatus comprising: a memory; a processor coupled to the memory, the processor being configured to execute a region extraction processing that includes extracting a person region from a plurality of images, the person region corresponding to a region of a person, the plurality of images including an image taken at a first time point and an image taken at a second time point after the first time point; execute an identification processing that includes identifying a first body part and a second body part in the person region from each image at the first time point and the second time point, the first body part being a left shoulder of the person in the person region, the second body part being a right shoulder of the person in the person region; execute a calculation processing that includes calculating a first moving amount and a second moving amount, the first moving amount indicating a moving amount between a position of the first body part at the first time point and a position of the first body part at the second time point, the second moving amount indicating a moving amount between a position of the second body part at the first time point and a position of the second body part at the second time point; and execute an estimation processing that includes: in response to detecting that the first moving amount is larger than the second moving amount, estimating body orientation of the person in the person region at the second time point by using first relation information and a difference between the first moving amount and the second moving amount, the first relation information indicating a relationship between the difference and a first change amount of the body orientation when the person rotates clockwise; and in response to detecting that the second moving amount is larger than the first moving amount, estimating the body orientation of the person in the person region at the second time point by using second relation information and the difference, the second relation information indicating a relationship between the difference and a second change amount of the body orientation when the person rotates counter-clockwise.
7. The apparatus according to claim 6 , wherein the first relation information and the second relation information are set for each body orientation at the first time point, the estimation processing is configured to: in response to detecting that the first moving amount is larger than the second moving amount, identify the first relation information corresponding to the estimated body orientation at the first time point; and in response to detecting that the second moving amount is larger than the first moving amount, identify the second relation information corresponding to the estimated body orientation at the first time point.
8. The apparatus according to claim 6 , wherein the identification processing is configured to use feature amounts at a plurality of feature points extracted from the person region in order to identify the first body part and the second body part.
9. The apparatus according to claim 8 , wherein the processor is configured to store, as first feature points associated with the first time point, second feature points identified from the person region at the second time point; and store the estimated body orientation at the second time as the body orientation at the first time.
10. The method according to claim 6 , wherein the region extraction processing is configured to extract a plurality of person regions in an image, wherein the identification processing is configured to identify, for each of the plurality of person regions, the first body part in the person region and the second body part in the person region, wherein the calculation processing is configured to calculate, for each of the plurality of person regions, the first moving amount and the second moving amount, and wherein the estimation processing is configured to estimate, for each of the plurality of person regions, the body orientation of the person in the person region at the second time.
11. A non-transitory computer-readable storage medium for storing a program which causes a processor to perform processing for estimating body orientation, the processing comprising: executing a region extraction processing that includes extracting a person region from a plurality of images, the person region corresponding to a region of a person, the plurality of images including an image taken at a first time point and an image taken at a second time point after the first time point; executing an identification processing that includes identifying a first body part and a second body part in the person region from each image at the first time point and the second time point, the first body part being a left shoulder of the person in the person region, the second body part being a right shoulder of the person in the person region; executing a calculation processing that includes calculating a first moving amount and a second moving amount, the first moving amount indicating a moving amount between a position of the first body part at the first time point and a position of the first body part at the second time point, the second moving amount indicating a moving amount between a position of the second body part at the first time point and a position of the second body part at the second time point; and executing an estimation processing that includes: in response to detecting that the first moving amount is larger than the second moving amount, estimating body orientation of the person in the person region at the second time point by using first relation information and a difference between the first moving amount and the second moving amount, the first relation information indicating a relationship between the difference and a first change amount of the body orientation when the person rotates clockwise; and in response to detecting that the second moving amount is larger than the first moving amount, estimating the body orientation of the person in the person region at the second time by using second relation information and the difference, the second relation information indicating a relationship between the difference and a second change amount of the body orientation when the person rotates counter-clockwise.
12. The non-transitory computer-readable storage medium according to claim 11 , wherein the first relation information and the second relation information are set for each body orientation at the first time point, the estimation processing is configured to: in response to detecting that the first moving amount is larger than the second moving amount, identify the first relation information corresponding to the estimated body orientation at the first time point; and in response to detecting that the second moving amount is larger than the first moving amount, identify the second relation information corresponding to the estimated body orientation at the first time point.
13. The non-transitory computer-readable storage medium according to claim 11 , wherein the identification processing is configured to use feature amounts at a plurality of feature points extracted from the person region in order to identify the first body part and the second body part.
14. The non-transitory computer-readable storage medium according to claim 13 , wherein the processing further includes storing, as first feature points associated with the first time point, second feature points identified from the person region at the second time point; and storing the estimated body orientation at the second time as the body orientation at the first time.
15. The non-transitory computer-readable storage medium according to claim 11 , wherein the region extraction processing is configured to extract a plurality of person regions in an image, wherein the identification processing is configured to identify, for each of the plurality of person regions, the first body part in the person region and the second body part in the person region, wherein the calculation processing is configured to calculate, for each of the plurality of person regions, the first moving amount and the second moving amount, and wherein the estimation processing is configured to estimate, for each of the plurality of person regions, the body orientation of the person in the person region at the second time.
This invention relates to computer vision systems for analyzing human movement in images, specifically estimating body orientation over time. The technology addresses the challenge of accurately tracking and analyzing multiple individuals in a scene, particularly their body movements and orientations, which is useful in applications like surveillance, sports analysis, and human behavior monitoring. The system processes images to extract multiple person regions, each representing an individual in the scene. For each person region, the system identifies two distinct body parts, such as the head and torso or arms and legs. The system then calculates the movement of these body parts between two different times, determining a first and second moving amount for each. Using these movement values, the system estimates the body orientation of each person at the later time. This allows for dynamic tracking of multiple individuals' postures and movements, even in crowded or complex environments. The approach improves upon prior methods by handling multiple subjects simultaneously and providing detailed movement and orientation data for each, which is critical for applications requiring precise human activity analysis. The system's ability to process multiple person regions independently ensures scalability and accuracy in diverse scenarios.
Unknown
February 2, 2021
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.